'Do Anything Now'¶
Shen, X., Chen, Z., Backes, M., Shen, Y., & Zhang, Y. (2024). 'Do Anything Now': Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models. Proceedings of the 2024 ACM SIGSAC Conference on Computer and Communications Security (CCS '24).
Cited by¶
1 citation across 1 artifact.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Community-Distributed Adversarial Learning
- In AI safety it is jailbreak adaptation against model safety filters — the originating instance, where a community curates a corpus of working prompts that outpace the model-update cycle.
This sourceAlso arXiv:2308.03825. Collects and analyzes 1,405 in-the-wild jailbreak prompts shared and refined across online communities, the originating instance of a community-curated bypass corpus outpacing the model safety-update cycle.
Supported in partVerified against the work's full text
Shen et al. document community-shared and curated jailbreak prompts and a cat-and-mouse game with vendors, but also show a late-2023 ChatGPT update neutralised most of them.
“Our results suggest that, despite the efforts from OpenAI, it is still difficult to identify and mitigate all jailbreak prompts.”
- In AI safety it is jailbreak adaptation against model safety filters — the originating instance, where a community curates a corpus of working prompts that outpace the model-update cycle.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Read against the text for 1 of 1 citation: 1 supported in part. Each verdict is shown under its citation below, with what in the work backs the sentence.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Registry ID ref:31efadf7477c · see in the full table